In a recent development. Scientists have developed AI software that can engineer proteins that may be useful for cancer treatment, vaccines, or even work as tools to draw carbon out of the air.
The research reported in the journal Science is led by Harvard University and University of Washington School of Medicine.
“The protein found in nature are amazing molecules, but the ones created can do much more, stated the senior author of the study.” The work demonstrates that ML can be used to design proteins with a wide range of functions.
Meanwhile, for several years, scientists have used computers attempting to create proteins. Some proteins such as synthetic binding proteins, and antibodies have been adapted into medicines to fight COVID-19. Some other engineered protein such as enzymes help in industrial processes. But a single protein molecule often comprises a large number of bonded atoms, which are difficult to engineer and study even with specialized scientific programming tools.
Inspired by the might of ML to create stories or even images from prompts, the team of scientists undertook to build similar software to design new protein. The idea does not change: Neural networks can be guided to see patterns in data. Once coached, it can be prompted and viewed if it can produce an effective solution. The results are often compelling, or even elegant, said the lead author of the study.
In fact, taking information from the Protein Data Bank, the team directed multiple neural networks which is a public database of very large number of protein structures from all kingdoms of life. The resultant neural networks have even surprised the scientists who created them. The team developed two approaches to design protein with two functions.